Machine learning-assisted design of inverted gradient multiring optical fibres for flat-top beam generation

This paper introduces a new method for designing inverted gradient multilayer optical fibres to generate flat-top laser beams with specific propagation features. Using numerical simulations in OptiFibre software, over 700 three-layer optical fibre configurations were modelled, and their output beam profiles were assessed based on beam flatness, edge steepness, and full width at half maximum (FWHM). A unique set of parameters describing the refractive index profile was proposed to improve the estimation of beam properties. The random forest algorithm was employed to predict beam characteristics based on fibre geometry, and a feature importance analysis identified key parameters that influence these characteristics. The model showed high prediction accuracy, particularly when the proposed descriptors were combined with layer refractive indices. Additionally, two multi-criteria optimisation methods - trade-off and weighted objectives - were used to find optimal fibre configurations that simultaneously maximise beam flatness, edge steepness, and FWHM. The findings offer practical guidelines for the inverse design of beam-shaping optical fibres and demonstrate the potential of machine learning to accelerate fibre design by reducing dependence on lengthy simulations.